Distributed Application Service Fault Management Using Bayesian Network

According to the features of the distributed application service fault management, we propose a hybrid fault propagation model for fault detection, which includes a multi-layer FPM model and a two-layer FPM model. And the diagnosis process is divided into two procedures: application service fault diagnosis and network service fault diagnosis. Because the observation of faults is uncertain, we map the fault diagnosis model to Bayesian network to carry out uncertainty reasoning. To improve the inference speed, we add the bucket elimination algorithm with minimum deficiency for better order. In addition, according to the sparse nature of the multi-layer FPM model graph, we use ancestral set to simplify the graph to improve the inference algorithm. As experiments shown, the optimized bucket elimination is improved a lot at speed.